. 2016 Nov 7;3(4):046002. doi: 10.1117/1.JMI.3.4.046002

Table 1.

Reported segmentation errors for prostate segmentation algorithms intended for use on T2w ER MRI.

Group	Method	Dataset size	Accuracy	Segmentation time
Our group¹⁷	Local appearance and shape model (semiautomatic)	42 (test and training)	Whole gland:MAD: $2.0 \pm 0.5 mm$ DSC: $82 % \pm 4 %$ Recall: $77 % \pm 9 %$ Precision: $88 % \pm 6 %$ $Δ V$ : $- 4.6 \pm 7.2 {cm}^{3}$	Operator interaction: $28 \pm 14 s$ . (across 10 images and 9 operators) Execution: $85 \pm 20 s$ . (across 42 images, one operator)
Cheng et al.²¹	Atlas-based (automatic)	100 (training) and 40 (test)	Whole gland:TP: 91.2%DSC: 87.6% $Δ V$ : 8.4%	NA
Liao et al.¹⁸	Multi-atlas-based (automatic)	66 (test) 9 (atlas)	Whole gland:MAD: $1.8 \pm 0.9 mm$ DSC: $88 % \pm 3 %$	Execution: 2.9 min
Toth and Madabhushi¹⁹	Active appearance model (semiautomatic)	108	Whole gland:MAD: $1.5 \pm 0.8 mm$ DSC: $88 % \pm 5 %$	Execution: 150 s
Vikal et al.²²	Shape model (semiautomatic)	3	Has not reported for whole gland	Execution: 23 s
Martin et al.²⁰	Atlas-based (semiautomatic)	1 (reference) 17 (test)	Whole gland:MAD: $3.4 \pm 2.0 mm$ Recall: $89 % \pm 6 %$ Precision: $78 % \pm 12 %$	NA

Note: MAD, mean absolute distance; DSC, Dice similarity coefficient; $Δ V$ , volume difference; TP, true positive.